Allele frequency estimation error vs. true allele frequency on simulated data.

Allele frequency estimation error vs. true allele frequency on simulated data. The dashed lines depict the standard deviation, solid line depicts the mean. The grey area shows the standard deviation of a corresponding binomial sampling, which marks the maximum theoretically achievable accuracy.

Allele frequency estimation error vs. true allele frequency on simulated data.

Allele frequency estimation error vs. true allele frequency on simulated data. The dashed lines depict the standard deviation, solid line depicts the mean. The grey area shows the standard deviation of a corresponding binomial sampling, which marks the maximum theoretically achievable accuracy.

For concordance analysis, we ran varlociraptor and the other callers on four independent sequencings of the same tumor/normal pair. Called variants were matched against each other, yielding a graph. Connected components in this graph of more than one node we counted as concordant calls. Left plot shows concordance vs minimum allele frequency. Right plot shows number of calls vs minimum allele frequency. The grey lines depict the theoretical expectation according to the model of neutral tumor evolution by Williams et al. See paper for details.

For concordance analysis, we ran varlociraptor and the other callers on four independent sequencings of the same tumor/normal pair. Called variants were matched against each other, yielding a graph. Connected components in this graph of more than one node we counted as concordant calls. Left plot shows concordance vs minimum allele frequency. Right plot shows number of calls vs minimum allele frequency. The grey lines depict the theoretical expectation according to the model of neutral tumor evolution by Williams et al. See paper for details.

Performance of FDR control on simulated data. Horizontal axis shows the applied FDR threshold, vertical axis shows the true FDR of the selected set of variants. Perfect FDR control would result in a line directly on the diagonal. Conservative control would lie in the area below the diagonal. Anything above the diagonal would be overly optimistic.

Performance of FDR control on simulated data. Horizontal axis shows the applied FDR threshold, vertical axis shows the true FDR of the selected set of variants. Perfect FDR control would result in a line directly on the diagonal. Conservative control would lie in the area below the diagonal. Anything above the diagonal would be overly optimistic.

Precision and recall on simulated data.

Precision and recall on simulated data.

Distribution of true and false positives depending on posterior probability for somatic variant in the tumor as given by varlociraptor. Desirable is a good separation between the two distributions.

Distribution of true and false positives depending on posterior probability for somatic variant in the tumor as given by varlociraptor. Desirable is a good separation between the two distributions.